loci 🧠

Two thousand years ago, orators stored their speeches in the rooms of a palace and walked through them to remember. loci does the same for your files.

Loci is the method behind every memory palace: place knowledge in locations, recall it by walking the path.

loci demo

A queryable "second brain" for the project docs, notes, and chat logs scattered across a dozen directories — and an MCP server so your AI agents can use it too.

Local files → heading-aware chunking → embeddings → hybrid retrieval (vector + BM25) → LLM answer with section-level citations. The index lives entirely on your machine; only embedding/chat calls go out, to any OpenAI-compatible API (Zhipu / DeepSeek / Kimi / OpenAI / …).

The thesis (from studying the 90k-star platforms and the graveyard of dead lightweight tools — see our competitive landscape study): don't build another chat app. Build the memory layer that every chat app can mount. Claude Desktop, Cursor, Cline, or any MCP host becomes this project's UI, for free.

Demo

Real session, indexed against the docs of minimax-h3-turing (paths shortened for display):

Code

$ python main.py search "what the 22G card can and cannot do" -k 3

[1] minimax-h3-turing/docs/en/01-hardware-limits.md > 01 · What a 2080Ti 22G Can and Cannot Do    (similarity 0.562)
[2] minimax-h3-turing/docs/en/02-w4a8-vs-w4a4.md > 02 · Quantization Measured > You Can Try Without 22G  (similarity 0.446)
[3] minimax-h3-turing/docs/en/01-hardware-limits.md > ... > 3. VRAM is just barely enough — manage it  (similarity 0.504)

$ python main.py ask "How should I choose between T8 aggressive mode and the final-render mode, and why?"

Answer:
* Drafts / preview / shot selection: use T8 aggressive mode — a 43% speedup
  (2.7 min/clip), and "a different picture of equal quality" is fine for picking shots.
* Final shots: use final-render mode (no T8). T8 makes the numerical trajectory
  fork, so re-running with the same seed produces a different clip — which breaks
  the reproducibility final outputs need.

[source: docs/en/08-t8-blockcache-4step.md > Practical Advice (4-step Turbo route)]
[source: docs/en/06-faq.md > 12. Cache-style accelerators break "same-seed re-runs"]

Hybrid retrieval means a Chinese query still finds the English doc (and vice versa) — keyword evidence (BM25) catches what embeddings miss, and every citation points at a section, not just a file.

Does hybrid actually help? (mini-eval, 10 bilingual queries)

Code

$ python scripts/eval_retrieval.py scripts/eval_cases.example.jsonl
vector-only: 9/10  →  hybrid: 10/10

Hybrid also fixed the #1 ranking on keyword-ish queries (e.g. "T8 block cache threshold speedup": vector put an FAQ first, hybrid puts the actual T8 writeup first). Run it against your own corpus with your own cases file.

Reranking: two providers

--rerank reorders the fused candidates for precision:

Provider	How	Cost
`llm` (default)	pointwise 0–3 relevance scoring by your chat model	one extra LLM call
`local`	cross-encoder, via `pip install 'loci[rerank]'`	~30–70 ms for 5 pairs on GPU — offline, free

bash

python main.py search "T8 speedup" --rerank          # provider from config
python main.py search "T8 speedup" --rerank local    # cross-encoder (BAAI/bge-reranker-base)

The local model downloads on first use (~1.1 GB; set HF_ENDPOINT=https://hf-mirror.com if HuggingFace is slow in your region). Measured on a 2080 Ti, bilingual query.

Office documents, PDF tables, web pages, org files, chat logs

PDFs: with the [pdf] extra, PyMuPDF4LLM extracts pages as markdown — tables come through as pipe rows (plain pypdf text is the fallback)
Word: with the [docx] extra, .docx paragraphs and table rows are indexed
HTML: .html / .htm pages become text with headings preserved (stdlib html.parser, zero dependencies — <meta charset> honored, script/style skipped)
org-mode: .org notes convert faithfully — #+TITLE becomes the h1 with *-sections nested under it, #+FILETAGS become searchable tags
Chat exports: drop a ChatGPT or Claude conversations.json into any source directory — it becomes one searchable document per conversation, tagged chatlog (search --tag chatlog scopes to chat history)

How it relates to Obsidian / your note app

It doesn't compete — the two layer up. Obsidian (or any editor) is the note-taking frontend; this is the cross-vault search engine: point sources at any directories (Obsidian vaults, project docs, chat exports) and query all of them at once — from your terminal, your scripts, or your AI agent via MCP. Obsidian-native details are understood: frontmatter tags: (filter with search --tag), [[wikilinks]] (walk the graph with links), code blocks are never cut mid-block, and one-line notes stay searchable.

How it works

mermaid

%%{init: {'theme':'base','themeVariables':{'background':'#000000','primaryColor':'#000000','primaryTextColor':'#00FF41','primaryBorderColor':'#00FF41','lineColor':'#00FF41','secondaryColor':'#001a00','tertiaryColor':'#000000','clusterBkg':'#000000','clusterBorder':'#00FF41','edgeLabelBackground':'#000000','fontSize':'14px','fontFamily':'trebuchet ms, verdana, arial, sans-serif'},'themeCSS':'.nodeLabel { color: #00FF41 !important; } .edgeLabel { background: #000 !important; color: #00FF41 !important; } .cluster-label { color: #00FF41 !important; }'}}%%
flowchart LR
    subgraph sources["📥 Your machine"]
        notes["Obsidian / markdown notes"]
        docs["PDF tables · docx · project docs"]
        chats["ChatGPT / Claude exports"]
        mem["memories/ — agent-written notes"]
        wikidir["wiki/ — consolidated pages"]
    end

    subgraph loci["🧠 loci — local index, nothing leaves the machine"]
        ingest["ingest / watch<br>loaders → chunker → embedder"]
        store[("ChromaDB<br>hybrid index")]
        retrieve["hybrid retrieval<br>vector + BM25 → RRF"]
        mcp["loci-mcp<br>8 tools · resources · prompts"]
    end

    subgraph hosts["🖥️ Your AI hosts"]
        ide["Claude Code · Qoder · Trae<br>Cursor · Cline"]
        desktop["Claude Desktop"]
        term["Terminal<br>search / ask / chat / wiki"]
    end

    api["☁️ OpenAI-compatible API<br>Zhipu / DeepSeek / Kimi / OpenAI<br>or 100% offline via Ollama"]

    sources --> ingest --> store
    mem -. auto-indexed .-> store
    wikidir -. auto-indexed .-> store
    store --> retrieve
    retrieve --> term
    retrieve --> mcp
    mcp <--> ide
    mcp <-.-> desktop
    retrieve -. "embedding + chat calls only" .-> api

The write path in one line: loaders → chunker (heading-aware split) → embedder → store (ChromaDB, persistent) — incremental, deduplicated by content hash.

Install & quick start

Requires Python 3.11+ (uses the stdlib tomllib).

bash

# option A: install from PyPI (adds `loci` and `loci-mcp` commands)
pip install "loci-rag[pdf,docx]"   # optional extras: PDF w/ tables, Word documents

# option B: zero-install quickstart
pip install -r requirements.txt

# 1. Configure: copy the example and fill in your values
cp config.example.toml config.toml

# 2. Ingest (incremental — deduplicated by content hash, safe to re-run)
loci ingest            # or: python main.py ingest

# 3. Ask
loci ask "what did I write about X?"

The workflow

mermaid

%%{init: {'theme':'base','themeVariables':{'background':'#000000','primaryColor':'#000000','primaryTextColor':'#00FF41','primaryBorderColor':'#00FF41','lineColor':'#00FF41','secondaryColor':'#001a00','tertiaryColor':'#000000','clusterBkg':'#000000','clusterBorder':'#00FF41','edgeLabelBackground':'#000000','fontSize':'14px','fontFamily':'trebuchet ms, verdana, arial, sans-serif'},'themeCSS':'.nodeLabel { color: #00FF41 !important; } .edgeLabel { background: #000 !important; color: #00FF41 !important; } .cluster-label { color: #00FF41 !important; }'}}%%
flowchart TD
    A["pip install loci-rag"] --> B["cp config.example.toml config.toml<br>fill API keys + source dirs"]
    B --> C["loci ingest — hybrid index built"]
    C --> D["loci watch — index stays fresh (optional)"]
    C --> E{"What do you need?"}
    E -->|"a synthesized answer"| F["loci ask --verify<br>claim-by-claim audit"]
    E -->|"raw excerpts to quote"| G["loci search --tag memory"]
    E -->|"back-and-forth"| H["loci chat"]
    E -->|"scattered notes on a topic"| I["loci wiki topic<br>consolidate into a wiki page"]
    F --> J["loci remember —<br>keep what you learned"]
    I --> J

Commands

Command	What it does
`ingest`	scan sources, index new/changed files, prune deleted ones (`--force` re-embeds everything)
`search "query"`	retrieval only — ranked excerpts with `path > section` breadcrumbs
`ask "question"`	retrieval + LLM answer with `[source: path > section]` citations
`ask "…" --verify`	additionally audit the answer claim-by-claim against the sources (✓ supported, ~ partial, ✗ unsupported)

Filter operators (combine freely, on search and ask):

Flag	Filters to
`--tag foo`	files whose frontmatter tags contain `foo`
`--in docs/en`	files whose path contains the substring
`--since 2026-08` / `--since 2026-08-15`	files modified on/after that date
`-e "exact phrase"`	chunks containing the exact phrase
`-k N`	return N hits (default 5)
`links "note"`	show the `[[wikilink]]` graph around a note — outbound and inbound
`chat`	multi-turn Q&A loop with conversation memory (`/clear`, `/exit`)
`watch`	keep the index current by polling sources (interval in `[watch]`)
`ask "…" --rewrite`	LLM-rewrite the query (keyword + cross-language variants) before retrieval
`feedback good\|bad`	rate the chunks used in the last ask; bad-rated chunks sink in future results
`wiki --suggest`	suggest wiki-worthy topics that don't have a page yet
`bench cases.jsonl`	retrieval benchmark: hit@k, vector-only vs hybrid
`sync push\|pull`	sync memories/wiki across machines via git ([sync] remote)
`serve-http`	HTTP REST API (search/ask/remember/stats) with Bearer auth
`graph build` / `graph show ENTITY`	knowledge graph over memories/wiki (LLM-extracted triples in graph.json)
`stats`	what's in the index: chunks per source, models, retrieval settings
`doctor`	health check: config, source dirs, embed/LLM endpoints, store (exit code 1 on failure — CI-friendly)
`python mcp_server.py`	MCP server over stdio (see below)

One memory, every IDE

Because every MCP host mounts the same loci server (same config.toml, same index), memory written from one tool is recalled from every other:

bash

# Claude Code
claude mcp add loci -- loci-mcp

jsonc

// Cursor / Cline / Qoder / Trae (mcpServers JSON — same shape everywhere)
{ "mcpServers": { "loci": { "command": "loci-mcp" } } }

Then, from any of them: "remember that the staging password rotates on Mondays" → brain_remember → later, from a different IDE: "when does the staging password rotate?" → answered, with the memory cited. Memories live as plain markdown in the memories directory (git-friendly, no lock-in) and are tagged memory, so loci search --tag memory scopes to them.

Cross-IDE tip: the default store / memories paths are relative to the directory loci is launched from. If your IDEs start in different project folders, point both at one absolute location in config.toml — e.g. store.path = "~/.loci/store" and memories.path = "~/.loci/memories" — and every IDE shares the exact same memory store.

mermaid

%%{init: {'theme':'base','themeVariables':{'background':'#000000','primaryColor':'#000000','primaryTextColor':'#00FF41','primaryBorderColor':'#00FF41','lineColor':'#00FF41','actorBkg':'#000000','actorBorder':'#00FF41','actorTextColor':'#00FF41','signalColor':'#00FF41','signalTextColor':'#00FF41','noteBkgColor':'#001a00','noteBorderColor':'#00FF41','activationBkgColor':'#001a00','edgeLabelBackground':'#000000','fontSize':'14px','fontFamily':'trebuchet ms, verdana, arial, sans-serif'},'themeCSS':'.messageText { fill: #00FF41 !important; } .actor { fill: #000 !important; stroke: #00FF41 !important; } text.actor { fill: #00FF41 !important; }'}}%%
sequenceDiagram
    participant CC as Claude Code
    participant L as loci-mcp
    participant S as ChromaDB (local)
    participant T as Trae / Qoder / any IDE
    CC->>L: brain_remember("deploy rotates Mondays")
    L->>S: write memory.md + embed + index
    Note over S: persists across sessions and IDEs
    T->>L: brain_search("password rotation")
    L->>S: hybrid retrieval
    L-->>T: cited answer — the memory is recalled

Ecosystem

loci-dsh — visual plugin for DeepSeek Harness: search, ask, quick-capture memories and watch index stats from a sidebar in the dsh web UI, talking to loci serve-http over local REST.

Mount it in any MCP host

Add to claude_desktop_config.json (Claude Desktop) or your MCP client's config:

config.json

{
  "mcpServers": {
    "loci": {
      "command": "python",
      "args": ["/path/to/loci/mcp_server.py"]
    }
  }
}

The server exposes three tools (zero dependencies beyond the core):

Tool	Purpose
`brain_search(query, k?, tag?, in?)`	ranked excerpts with breadcrumbs
`brain_ask(question, verify?)`	grounded answer with citations; `verify=true` adds a claim-by-claim audit
`brain_links(note)`	outbound/inbound `[[wikilink]]` graph around a note
`brain_stats()`	index overview (chunks per source)
`brain_graph(entity?)`	knowledge-graph relations for an entity (omit for hub entities)
`brain_remember(text, title?, tags?)`	write a memory — durable, shared across sessions and IDEs
`brain_forget(query)`	soft-delete matching memories (they go to a `.trash` folder)
`brain_wiki(topic)`	memory consolidation — distill the index into a curated wiki page about a topic
`brain_ingest(force?)`	incremental re-index

Beyond tools, the server speaks the full protocol:

Resources — resources/list exposes brain://stats plus one brain://note/… resource per indexed file (raw markdown via resources/read)
Prompts — three ready-made templates: brain-briefing, study-plan, contradiction-check; hosts render them with your topic pre-filled

Fully offline with Ollama

The index is local by design — and the embedding/chat calls can be too. Any OpenAI-compatible server works; Ollama is verified end-to-end:

toml

[llm]
base_url = "http://localhost:11434/v1"
api_key = "ollama"          # any non-empty placeholder
model = "qwen2.5:0.5b"

[embed]
base_url = "http://localhost:11434/v1"
api_key = "ollama"
model = "all-minilm"

With this config, ingest / search / ask make zero cloud calls. Swap in a bigger local chat model for better answers — the pipeline is model-agnostic.

Configuration

Key	Meaning
`[llm]`	base_url / api_key / model — any OpenAI-compatible endpoint
`[embed]`	same; the model must be an embedding model (e.g. `embedding-3`)
`[[sources]]`	document directories, scanned recursively for `.md` / `.txt` / `.html` / `.org` (plus `.pdf`/`.docx`/images with the matching extras)
`[[sources]] chunk_size` / `chunk_overlap`	optional per-directory chunking override — wins over the global `[chunk]` block
`[chunk]`	chunking params (default 800 chars / 100 overlap)
`[top_k]`	number of hits per search (default 5)
`[retrieval]`	`hybrid` (vector+BM25 fusion, default on), `rrf_k`, `rerank` (LLM reranking, default off)
`[watch]`	poll `interval` seconds

API keys can also come from the environment variables BRAIN_LLM_API_KEY / BRAIN_EMBED_API_KEY (these override the config file).

Development

bash

git clone https://github.com/IvenKooLab/loci && cd loci
pip install -e ".[pdf,docx]"        # editable install for hacking on loci
pip install -r requirements-dev.txt
pytest                              # fully offline, no API keys needed

See CONTRIBUTING.md for the ground rules (no frameworks, tests stay offline, citations are sacred).

Design decisions

~300 lines of core, no LangChain — every stage is readable, hackable, and learnable. The whole engine fits in one sitting.
MCP-first — the agent ecosystem is the UI layer. No web app to maintain.
Hybrid retrieval on by default — vector search fused with a native ~60-line BM25 (CJK-aware tokenizer) via Reciprocal Rank Fusion.
Citations always, with breadcrumbs — path > section, so claims are verifiable at a glance.
Robust, inspectable indexing — defensive loaders (skip what can't be parsed, never hang), content-hash incrementality, real pruning, stats and doctor so the index is never a black box.
Tiny notes stay searchable — no minimum-chunk filter; a one-line note is still indexed (a lesson from watching other tools drop or choke on them).
Keys never in code — config.toml (gitignored) or env vars.

Where it sits

	loci	AnythingLLM (65k★)	Khoj (37k★)	RAGFlow (90k★)
Positioning	personal retrieval backend + MCP	all-in-one chat platform	self-hosted AI assistant	enterprise RAG engine
Footprint	2 runtime deps, no Docker	desktop app / Docker	Django server + workers	Docker, DeepDoc models
UI	your terminal & your agents	built-in web/desktop	web + Obsidian/Emacs	web
MCP server	✅ native	consumer	—	—
Hackable core	✅ ~300 lines	❌	❌	❌
Multi-user	by design, no	✅	✅	✅

(Full data and reasoning: competitive landscape study.)

Roadmap

See docs/roadmap.md — reranking, GraphRAG experiments, more loaders.

License

MIT